A Power Capping Approach for HPC System Demand Response

Event Sponsor: 
Mathematics and Computer Science Division
Start Date: 
Jul 27 2017 - 10:30am
Building/Room: 
Building 240/Room 4301
Location: 
Argonne National Laboratory
Speaker(s): 
Kishwar Ahmed
Speaker(s) Title: 
Florida International University
Host: 
Kazutomo Yoshii

Abstract: Demand response programs ensure power grid transmission stability by intelligently reducing energy during electricity peak demand times or emergency incidents. Being a massive energy consumer, high-performance computing (HPC) system can be considered an ideal participant in demand response programs. In this talk, we discuss possibility of HPC system’s participation in demand response program, and propose a job scheduling and resource provisioning scheme for such participation. Since modern processors provide hardware-level power capping and opportunistically adjusts its frequency based on thermal and energy constraints (e.g., Intel's Turbo Boost technology), we exploit such capability (i.e., power capping) to enable HPC energy reduction during demand response times. Our proposed demand response scheme includes detailed models for various subcomponents (e.g., processor, memory, cooling) in an HPC system. We develop a job scheduling simulator to show effectiveness of our HPC system demand response model. We evaluate the effectiveness of our demand response job scheduling and resource allocation scheme by applying the characteristics of real-world HPC scientific applications. Our demand response model can achieve both power stability and energy saving for HPC system, while ensuring system-wide power-bound constraint.  

Bio: Kishwar Ahmed is a Ph.D. candidate in the School of Computing and Information Sciences at Florida International University in USA. He received his B.Sc. in Computer Science and Engineering from Bangladesh University of Engineering and Technology in Bangladesh and M.Sc. in Computer Science from Florida International University in USA. His research interests include energy-aware scheduling, performance prediction, parallel and distributed computing, and optimization.